Comparing sensor-based activity features of frailty across Fried Frailty Phenotype and deficit-accumulation frailty index

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Background and Objectives Wearable sensors offer a low-burden approach to assessing frailty in free-living settings, but how sensor-derived features relate to different frailty definitions is unclear. This study compared thigh-worn sensor-derived activity features associated with pre-frail/frail status under the Fried Frailty Phenotype (FFP) and the Comprehensive Geriatric Assessment–Based Frailty Index (CGA-FI). Research Design and Methods We enrolled 44 older adults (mean age 82.3 years, 84% women) from senior housing communities. Participants wore a thigh-mounted ActivPAL accelerometer for 10 days and completed FFP and CGA-FI assessments. Thirty-four activity features were extracted across activity volume, postural transitions, gait, sedentary behavior, and diurnal variability domains. Point-biserial correlations compared feature associations with pre-frail/frail status under each frailty definition. Ridge, lasso, and random forest classifiers were evaluated secondarily. Results Lower activity-volume features were associated with pre-frail/frail status under both definitions. Lower cadence was more strongly associated with FFP-defined pre-frail/frail status, whereas sedentary accumulation and lower inter-daily stability showed stronger associations with CGA-FI-defined pre-frail/frail status. Ridge logistic regression showed moderate discrimination of robust versus pre-frail/frail status (FFP: AUC 0.81 and F1 0.78; CGA-FI: AUC 0.79 and F1 0.77). Discussion and Implications Thigh-worn accelerometry captured clinically plausible activity patterns associated with pre-frail/frail status under both definitions. Some individual feature associations differed in interpretable ways across definitions. Secondary analyses suggested that discriminatory information was largely overlapping in this sample. These findings support frailty-definition-aware interpretation of sensor-derived measures but do not imply the need for separate definition-specific multivariable models. Further evaluation of sensor-based frailty screening approaches is recommended.